It is often desired that an input image from an input device such as scanner or digital camera can be reproduced with high quality. To produce a digital quality with high quality, appropriate image process needs to be performed on input image. With the current digital image processing technology, an image of high quality can be generated.
However, problems exist in the prior art image processing technology. In general, an input image can be divided into a character portion, a photographic portion, and a dot portion. When a dot image is processed using a filtering method, due to the influence of sampling rate and pixel, the input image often blurs. On the other hand, when a file containing a dot image, a character image and a photographic image is processed, the produced image often contains undesired dots in its character portion, and the edge of the characters become unclear.
Edge detection is very important in digital image processing. Generally, edges in an image can show the shape of objects in an image, which is important both for displaying the objects and for analyzing the image. Thus, edge detection is key for solving many complex questions in the field of digital image processing.
For example, Japanese Laid-Open Patent Publication No. 6-150059 discloses an image separating method and a separating apparatus, in which the separating precision of an image area is improved by determining the concentrated portion of the lattices and lines in the image as a character portion. In this publication, an image is divided into blocks each containing a plurality of pixels, and the type of the image is recognized by pattern matching; a plurality of reference image patterns are preset, and the type of an input image is determined by comparing the image with the reference patterns.
In addition, Japanese Laid-Open Patent Publication No. 8-149297 discloses a method for determining whether a pixel to be observed is contained in a dot image portion, wherein the periodicity of the image data of lines in the dot image is detected by using the periodicity of the image data of the dot image.
Moreover, Japanese Laid-Open Patent Publication No. 61-194968 discloses a method for recognizing a dot image area, wherein whether the pixel to be observed is in the portion of the dot image area or not is determined by measuring the pixels in blocks having different signal levels in a main scanning direction and a sub-scanning direction.
In a gray-level image, an edge is consisted of the pixels that appear at positions where the gray-level varies greatly. Such a variation of gray-level can be detected using a derivative method. The book “Image Segmentation”, by Zhang, Yujin (Scientific Book Series in Images and Graphics, Science Press, 2000, ISBN 7-03-007241-3) discusses some of the most frequently used methods for detecting an edge in an image using 1st order derivative, such as Roberts operator, Sobel operator, Prewitt operator, etc. If the calculated derivative of the gray-level of the image at a pixel is greater than a preset threshold (an image edge detecting threshold), then the pixel is determined as an edge pixel of the image.
U.S. Pat. No. 6,192,153B1 discloses an image processing method and apparatus. It performs process for each pixel by precisely identifying the type of each image area of the image to be detected. The image process apparatus of U.S. Pat. No. 6,192,153B1, comprises an image edge detecting means for detecting pixels at an edge in the image by calculating the product of each image signal in an area and a filter factor, wherein the area contains a target pixel and a plurality of pixels surrounding the target pixel, and a labeling means for labeling an image area containing the target pixel in accordance with the result of the edge detecting. The image processing method of U.S. Pat. No. 6,192,153B1 comprises the following step: inputting the digital image signal; reversing the digital image; detecting edges in the image; classifying each pixel into a photographic pixel and a non-photographic pixel according to the characteristics of each pixel; classifying the non-photographic pixel; and, smoothing the photographic pixel so as to output the obtained image signal.
U.S. Pat. No. 5,583,659 discloses a multi-windowing technique for thresholding an image using local image properties, which uses features such as luminance variations of a local image and gradient variation of image pixels and reduces the noise of the image and lowers the blurring at edges in the image using multi-windowing technique. In the method of U.S. Pat. No. 5,583,659, an edge in the image is detected using Sobel operator and a predetermined fixed edge criterion threshold.
In practice, edge criterion threshold needs to be adjusted based on the difference between images to be processed, so as to obtain good edge detection result of both extracting edges as completely as possible and avoiding producing too much noise in the extracted image. In the prior art, as in the above-mentioned two US patents, fixed threshold is used for edge criterion, so satisfactory result cannot be obtained.